Subroutines vs Coroutines vs Generators

Coroutines vs Subroutines

Subroutines:

• invoked once and executes until it completed.
• can be translated to a coroutine which does not call yield.

Coroutines:

• a more generalized form of subroutines.
• can be entered, exited, and resumed at many different points.
• can pause execution and yield control back to the caller or another coroutine. The caller can then resume the coroutine when appropriate. (i.e. it may suspend its execution in one thread and resume in another one. A coroutine is not bound to any particular thread.)
• used for cooperative multitasking and are often compared to fibers, lightweight threads and green threads

Coroutines vs Generators

Generators: a.k.a. "semicoroutines": a more limited form of coroutine that may only yield to its caller. Primarily used to simplify the writing of iterators.

Both can yield multiple times, suspending their execution and allowing re-entry at multiple entry points.

Key difference:

• Generators: can only yield to its caller. Cannot specify a coroutine to jump to in the yield statement.
• Coroutines: can control where execution continues immediately after they yield.

Coroutines vs Threads

• threads: scheduled by the operating systems.
• coroutines: scheduled by the users. (A.k.a. lightweight threads or green threads.)

Coroutines in Languages

• C++: coroutines introduced in C++20.
• Python:
  • https://www.python.org/dev/peps/pep-0492/
  • async and await
• Kotlin: in an extension lib kotlinx.coroutines. (NO keywords and NOT in the standard library)
  • launch, delay, suspend fun
• Java: Project Loom.

Coroutines vs goroutines

• goroutines imply parallelism; coroutines in general do not.
• goroutines communicate via channels; coroutines communicate via yield and resume operations.

When to use Coroutines

• data structure iterators.
• event-driven code without the inversion of control.
• cooperative multitasking.
• concurrency frameworks such as actors, async-await and dataflow networks.
• expressing asynchrony, and better composition of asynchronous systems.
• capturing continuations.
• expressing backtracking algorithms.
• AI agents such as behavior trees.